Fix Open WebUI connection and sign-in problems

The two symptoms on this page, a sign-in screen that keeps coming back and a model list that stays empty, have different causes, and on the DeployGPU template the first one has a history.

Open WebUI: Server Connection Error
Connection error: Cannot connect to host ...

The first line is the heading of the troubleshooting file in Open WebUI's repository, and the second is what its server log prints, according to its connection troubleshooting page (both checked 1 October 2026). Sign-in loops show no error at all, only the same page again.

The app.deploygpu.ai link bounces you back to sign-in

The private link opens only for the DeployGPU account that launched the deployment, signed in to DeployGPU in a browser. Open WebUI also sends its own Authorization: Bearer header with its requests, and until 24 September 2026 the DeployGPU link gate rejected them, so users were sent back to sign-in in a loop. The gate logged 34 rejected requests on 22 September, and we fixed it on 24 September. It has logged none since.

If you still loop, or the page shows 403, work through three checks. Sign in to DeployGPU with the account that launched the deployment, since any other account is refused. Clear the cookies for app.deploygpu.ai and open the app again from the dashboard. Confirm the deployment still exists, because the link stops working the moment a deployment stops, and a VM the provider loses is stopped by the 5-minute health check without an email.

The link also does not accept API keys or Bearer tokens, so a script or an OpenAI client pointed at it only sees the sign-in redirect. For code, tunnel to the app port with the user, host and port from the dashboard's SSH command:

ssh -N -L 8080:127.0.0.1:8080 ubuntu@YOUR_VM_IP -p 22

Open WebUI has no default password

There is none. According to Open WebUI's quick start (checked 1 October 2026), the first account you create is the administrator, and sign-up then switches itself off unless the admin turns on New Sign Ups. On the template, that account exists only on this VM, so every new deployment asks you to create it again. The DeployGPU sign-in is a separate step and not an Open WebUI password.

If you forgot the password you chose, Open WebUI's documented reset assumes the database sits in a Docker volume named open-webui. The template starts its container without a volume, so that recipe does not apply as written. When the deployment holds nothing you need, stop it and launch a fresh one, where the first account is the admin again.

An empty model list means no ollama pull yet

The template preloads no models. Ask Ollama what it has:

sudo docker exec deploygpu-workload ollama ls

An empty table means nothing is pulled. Type a tag such as gpt-oss:20b into the model selector and let Open WebUI offer to download it, or pull it over SSH:

sudo docker exec -it deploygpu-workload ollama pull gpt-oss:20b

The Ollama URL no longer matches localhost:11434

DeployGPU sets OLLAMA_BASE_URL=http://localhost:11434, and Ollama runs inside the same container, so localhost is correct on the template. If you see Server Connection Error, first ask Ollama directly with the command above. A table means Ollama is answering. An error means the server inside the container is down, so read sudo docker ps and sudo docker logs --tail 50 deploygpu-workload.

If Ollama answers but Open WebUI still cannot reach it, open Settings, Admin, Connections and check the Ollama URL. The connection troubleshooting page says a saved connection takes precedence over the environment variable. Set it back to http://localhost:11434 and refresh the connection.

Outside the template, localhost means the container

If you run Open WebUI in Docker yourself with Ollama on the VM or in another container, localhost inside the Open WebUI container is the container, not the VM. Open WebUI's documentation points the connection at http://host.docker.internal:11434, with --add-host=host.docker.internal:host-gateway on the container, or runs the container with --network=host and OLLAMA_BASE_URL=http://127.0.0.1:11434. The first needs Ollama to listen beyond 127.0.0.1, which on a DeployGPU VM puts it on the internet, because DeployGPU does not firewall VMs. So I use host networking. Open WebUI's start script binds HOST to 0.0.0.0 unless you set it, so I set it to 127.0.0.1:

sudo docker run -d --network=host -v open-webui:/app/backend/data \
  -e OLLAMA_BASE_URL=http://127.0.0.1:11434 -e HOST=127.0.0.1 \
  --name open-webui --restart always ghcr.io/open-webui/open-webui:main

Open WebUI then listens on port 8080 of the VM, which the SSH tunnel above reaches.

When to contact support

Contact support@deploygpu.ai if the deployment is Running, you are signed in with the launching account, the cookies are cleared, and the link still loops, or if sudo docker ps shows the container restarting. Send the deployment ID and the time of your last attempt.